Skip to main content

How To Use Keras Trained CNN Models

Introduction


Keras is a popular deep learning api. It can run on top of Tensorflow, CNTK and Theano frameworks. Keras provides an easy to use interface which makes deep learning practice straight forward. It is widely used thus resources are easily accessible.

Objective

This article aims to give an introductory information about using a Keras trained CNN model for inference. This article does not contain information about CNN training.

Audience

This article assumes introductory information about python and Convolutional Neural Networks. For those who lack information may first begin with information from following resources.

Software Installation

Keras is a high level API. It requires a back-end framework to be installed. In this article, Tensorflow is used. Keras can transparently select CPU or GPU for processing. If use of GPU is desired, assuming presence of a  proper graphics card with a decent GPU, relevant drivers needs to be installed.

Installation is not a simple procedure. Prepare a Ubuntu System for Deep Learning can be read for installation details.

Trained Models

Training a CNN model requires specialization, a lot of data and decent hardware. Transfer learning may simplify those requirements but it is not in the scope of this article.

Keras provides already trained models. Trained models and information about how to use them can be found in Keras Applications. Those models are trained using Imagenet dataset.

Additional models can be found in my GitHub page which are created as part of my emotion recognition study. Model files can be found at deep-emotion-recognition repository. Those models are trained using FER-13 dataset which contains 7 emotions. Rest of the article uses emotion recognition models from my GitHub page.

Application Code 

Processing Pipeline

This code is pretty straight forward. For loading a modal a load_model utility method is used. For loading images image generator provided by Keras is used. Please not that 1 is used as batch size. This is because for some reason using batch sizes other than 1 resulted in slightly different validation results for the same model at consecutive executions which is not acceptable.

For dataset either original dataset can be downloaded from original Kaggle page or from repository under dataset directory. Also note that the application only uses images found under Val directory.

Additional emotion datasets can be used. Some example datasets are:

Conclusion

We successfully loaded and evaluated a trained CNN model using Keras library. For full code listings you may check my GitHub source code repository.
Labels:

Comments

Post a Comment

Popular posts from this blog

Obfuscating Spring Boot Projects Using Maven Proguard Plugin

Introduction Obfuscation is the act of reorganizing bytecode such that it becomes hard to decompile. Many developers rely on obfuscation to save their sensitive code from undesired eyes. Publishing jars without obfuscation may hinder competitiveness because rivals may take advantage of easily decompilable nature of java binaries. Objective Spring Boot applications make use of public interfaces, annotations which makes applications harder to obfuscate. Additionally, maven Spring Boot plugin creates a fat jar which contains all dependent jars. It is not viable to obfuscate the whole fat jar. Thus obfuscating Spring Boot applications is different than obfuscating regular java applications and requires a suitable strategy. Audience Those who use Spring Boot and Maven and wish to obfuscate their application using Proguard are the target audience for this article. Sample Application As the sample application, I will use elastic search synch application from my GitHub repository.
4 comments
Read more

Hadoop Installation Document - Standalone Mode

This document shows my experience on following apache document titled “Hadoop:Setting up a Single Node Cluster”[1] which is for Hadoop version 3.0.0-Alpha2 [2]. A. Prepare the guest environment Install VirtualBox. Create a virtual 64 bit Linux machine. Name it “ubuntul_hadoop_master”. Give it 500MB memory. Create a VMDK disc which is dynamically allocated up to 30GB. In network settings in first tab you should see Adapter 1 enabled and attached to “NAT”. In second table enable adapter 2 and attach to “Host Only Adaptor”. First adapter is required for internet connection. Second one is required for letting outside connect to a guest service. In storage settings, attach a Linux iso file to IDE channel. Use any distribution you like. Because of small installation size, I choose minimal Ubuntu iso [1]. In package selection menu, I only left standard packages selected.  Login to system.  Setup JDK. $ sudo apt-get install openjdk-8-jdk Install ssh and pdsh, if not already i
3 comments
Read more

Java: Cost of Volatile Variables

Introduction Use of volatile variables is common among Java developers as a way of implicit synchronization. JIT compilers may reorder program execution to increase performance. Java memory model[1] constraints reordering of volatile variables. Thus volatile variable access should has a cost which is different than a non-volatile variable access. This article will not discuss technical details on use of volatile variables. Performance impact of volatile variables is explored by using a test application. Objective Exploring volatile variable costs and comparing with alternative approaches. Audience This article is written for developers who seek to have a view about cost of volatile variables. Test Configuration Test application runs read and write actions on java variables. A non volatile primitive integer, a volatile primitive integer and an AtomicInteger is tested. Non-volatile primitive integer access is controlled with ReentrantLock and ReentrantReadWriteLock  to compa
Post a Comment
Read more